ACPI is funny (well sort of)

kha · n00b Joined: 19 Mar 2003 Posts: 41

I have been playing around with quite a lot of different ACPI acces system (/proc/ACPI, /proc/cpufreq, /system/class) and here are the results :

/proc/ACPI won't react to anything, I can echo all I want, nothing moves.
I have tried echo -n 1 > throttling and echo -n 1:0 > limits but the speed won't move by an inch, my commands are ignored.

Conerning the other modes, then it is funny : I can't actually change the speed of my laptop but I can make the kernel think I did. The system clock will then go crazy and even the slightest touch of a key will make the character appears 5 times on the screen...

Does somebody have these kinds of symptoms ?

Kha
_________________
Digital is perfect, but humans keeps on being analogic.
Viagra Is A Great Reccursive Acronym (can't use it though :( )

pilla · Posted: Thu Mar 27, 2003 12:47 pm Post subject:

could you give more information about your notebook? Different brands and models have differents levels of ACPI support.
_________________
"I'm just very selective about the reality I choose to accept." -- Calvin

kha · n00b Joined: 19 Mar 2003 Posts: 41

My laptop is a Vaio PCG-GRZ660. The bios is a Phoenix (ARGH!) no way to switch to no pnp or to tweak anything (I can turn of the par port if I want though, and change the date too...). The processor is a P4 2.4 Ghz(not a mobile version though, so performance management is not present).

Appart form this little issue everything seems to be working fine.
In fact I already found quite a lot of way to save power, using a mix of HDPARM, spicctrl and iddle calls. Further more the screen is blanked after a while (even though I don't know how it is done, definitly not an ACPI event ).
So If i am not using my laptop my battery will last around 6 hours (against 2.5 when using it).

Funny thing is that pluging and unpluging the AC adaptator will trigger a CPU event, but will leave it at full speed (processor CPU0 000000080 000000000)

But my point is not really to get help, (even though it would be great if someone could actually help me), but to know if this problem happens on other PC/LAPTOP. I am trying to write a prog to manage ACPI through a GUI, so that is why I would like to know.

Kha
_________________
Digital is perfect, but humans keeps on being analogic.
Viagra Is A Great Reccursive Acronym (can't use it though :( )

kha · n00b Joined: 19 Mar 2003 Posts: 41

OK, so it was really a bug :) there was a problem with the throttling writting system in /proc/acpi. It is now fixed as of 2.5.66 and perfectly working.

Basically any writting operation aimed at throttling the CPU would fail, but the kernel would reccord it as OK and therefore adapt its speed accordingly, leading to the frenzy of the system.

Now it is fixed and working. Il will test the 2.5.66 completly and try to write a gui to manage the ACPI. But I can make my processor go as slow as 300 Mghz (is a 2.4 Ghz) without any problem.

Now I can finally say that everything is working perfectly in my GRZ660.
(And I won't have to recompile my dsdt, phew....)

Kha
_________________
Digital is perfect, but humans keeps on being analogic.
Viagra Is A Great Reccursive Acronym (can't use it though :( )

kha · n00b Joined: 19 Mar 2003 Posts: 41

Ok, here are the results. I can definitly save quite a lot of power by throttling down the CPU, but not as much as I expected. It looks like the best way to save power is to decrease the brightness of my LCD screen and to use hdparm to set the sleeping mode fast.

Since the 2.5.66 kernel is quite unstable for me (USB not working one time out of 5, strange errors and unexpected crashes..) I will quit using it for now. So I am back to my good old 2.5.65 who never failed me except for the buggy throttle implementation.

Kha
_________________
Digital is perfect, but humans keeps on being analogic.
Viagra Is A Great Reccursive Acronym (can't use it though :( )

abclements · Posted: Mon Mar 31, 2003 11:57 pm Post subject: Power Drain

Without a doubt the screen draws THE most power on a laptop. With the screen turned off I can get about 2.5 hours of my laptop doing whatever processing task I need. with the screen on full, it lasts about 40 min.
_________________
“Those who grasp at perceptions and views
go about butting their heads in the world.”
~The Buddha

magnet · Guru Joined: 16 Mar 2003 Posts: 582 Location: france

I had the same problem with my /proc/acpi , I finally switched to sysfs

if you still use the deprecated /proc/processor/ thing, there is a usefull tool called autospeedstep, that automagically switch your cpu speed, depending on the load.
I submitted a ebuild for it.

_________________
every step aim at glory.

kha · n00b Joined: 19 Mar 2003 Posts: 41

magnet · Guru Joined: 16 Mar 2003 Posts: 582 Location: france

hello,

I"m new to cpu throttling, I thinked that sysfs was the right was to do things.
what do you advice ? ( I have the clock weirdness problem) how should I change my cpu clock
I own a laptop with a p4 and I'm running a kernel 2.5.65-mm3. :roll:

_________________
every step aim at glory.

kha · n00b Joined: 19 Mar 2003 Posts: 41

Right now all kernel except 2.5.66 are buggy. Which means that tweaking the throttling in other kernel will result wether in nothing (good case) or time panic (bad case).

The only good thing to do under new kernels (4.19+ or 2.5.54+) is echo -n in /proc/acpi. All other methods are wether deprecated or dangerous.
Sysfs (/system/class/cpu/...) can report a success while a failure occured, leading to the clock madness, I tend not to use it for now, even though it will probably be the standard when everything is fixed.

the best way to do things is to use ACPI limit (/proc/acpi/processor/CPUx/limit) it is the safe and official way to do things right (from the acpi standard point of view). simply : echo -n p:t > limit
with p=predefine performance point, and t=throttling mode.
Globally echo -n 0:0 > limit will trigger the best speed available (t=0) when in performance mode (p=0). In the same way echo -n 1:7 > limit will trigger the worst speed (t=7) when in power-save mode (p=1).

In PC like mine which does not have a powersave mode (no performance management) everything is considered to be in p=0 mode.

By doing this you will actually update the user entry in limit, and if it is ok the actual state will conform. The good point is that if there is an alert (thermal for example) your entry will be ignored(that is in theory, actually almost nobody use this function).

You can also echo the speed you want directly into throttling, that's not the proper way of doing thing, and if one day a manufacturer conforms to ACPI standard, you might actually override some processor securities.

Once agian all of the above concerns throttling, not performance management and therefore works only with 2.5.66 kernel (and probably with a good set of ACPI patches which I am too lazy to investigate about right now).

Hope that helped
Kha
(au fait, je suis francais aussi)
_________________
Digital is perfect, but humans keeps on being analogic.
Viagra Is A Great Reccursive Acronym (can't use it though :( )